Content uploaded by Robert East
Author content
All content in this area was uploaded by Robert East
Content may be subject to copyright.
Measuring the impact of positive and negative word of mouth on brand
purchase probability
Robert East
a,
⁎
, Kathy Hammond
b,1
, Wendy Lomax
c,2
a
Kingston Business School, Kingston, KT2 7LB, UK
b
Duke Corporate Education, 165 Fleet St, London, EC4A 2DY, UK
c
Kingston Business School, Kingston, KT2 7LB, UK
ABSTRACTARTICLE INFO
Article history:
First received in February 22, 2008 and was
under review for 5 months
Keywords:
Word of mouth
Impact
Brand commitment
Familiarity
NPS
Using two methods, three measures, and data covering a large number of categories, we present findings on
the respondent-assessed impact of positive and negative word of mouth (PWOM, NWOM) on brand purchase
probability.
For familiar brands, we find that:
1. The impact of PWOM is generally greater than NWOM. The pre-WOM probability of purchase tends
to be below 0.5, which gives more latitude for PWOM to increase purchase probability than for NWOM to
reduce it.
2. The impact of both PWOM and NWOM is strongly related to the pre-WOM probability of purchase,
the strength of expression of the WOM, and whether the WOM is about the respondent's preferred brand.
3. PWOM and NWOM appear to be similar forms of advice-giving behavior, except for their opposed
effects on choice.
4. Respondents resist NWOM on brands they are very likely to choose, and resist PWOM on brands
they are very unlikely to choose.
In the Discussion section, we show how our methods could be used to construct a word-of-mouth metric.
© 2008 Elsevier B.V. All rights reserved.
1. Introduction
1.1. Defining the field
Word of mouth (WOM) is informal advice passed between
consumers. It is usually interactive, swift, and lacking in commercial
bias. WOM is a powerful influence on consumer behavior. Keaveney
(1995) noted that 50% of service provider replacements were found in
this way. WOM may be positive (PWOM), encouraging brand choice,
or negative (NWOM), discouraging brand choice.
Brand purchase probability will be affected by the relative inci-
dence of PWOM and NWOM about the brand and also by the relative
impact of instances of PWOM and NWOM. Here, we are concerned
with the impact of PWOM compared with NWOM. There is little
evidence on this matter, which may relate to the difficulty of making
accurate measurements in this field. Below, we review this measure-
ment problem.
WOM can affect the adoption of new categories and the choice of
brands in mature categories. In product adoption research, interest
falls on the few initial users of products whose advice to non-users
may decide the success or failure of a new product. In mature
categories, which are our research focus, changes occur mainly as
switching between brands and interest falls on users of the category,
who may be a majority of the population when categories such as cell
phones and restaurants are studied. Among users of mature
categories, WOM acts within a framework of acquired consumer
beliefs, preferences, habits, and commercial influences that may
constrain or assist response to the advice.
Research on the role of WOM in brand switching is required for
three reasons. First, WOM is often the major reason for brand choice,
but we do not yet understand how PWOM and NWOM contribute to
this influence. Second, some groups are more responsive to WOM than
others, and we show how segments with different probabilities of
purchase will respond differently to PWOM and NWOM. Third,
Reichheld's (2003) Net Promoter Score has performed poorly as a
predictor of brand/company performance. Our work provides some
explanation for this failure, and our methods may be used to develop a
better WOM metric.
Intern. J. of Research in Marketing 25 (2008) 215–224
⁎ Corresponding author. Tel.: +44 20 8547 2000x65563.
E-mail addresses: R.East@kingston.ac.uk (R. East), Kathy.Hammond@dukece.com
(K. Hammond), W.Lomax@kingston.ac.uk (W. Lomax).
1
Tel.: +44 20 7936 6140.
2
Tel.: +44 20 8547 7464.
0167-8116/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijresmar.2008.04.001
Contents lists available at ScienceDirect
Intern. J. of Research in Marketing
journal homepage: www.elsevier.com/locate/ijresmar
1.2. Difficultly in studying word of mouth
Although consumers often attribute their brand choice to WOM, it
is difficult to observe cases where advice affects brand choice since
WOM about a specific category is relatively uncommon and any effect
is often delayed. When evidence is scarce, too much weight may be
given to the limited research that is available. One solitary field study
by Arndt (1967) is often cited. Arndt found that NWOM had twice as
much impact on purchase as PWOM. However, he studied only one
brand, and systematic research should be based on all the brands in a
category and should include a range of categories. In addition,
although the category was familiar, Arndt used a new brand about
which there could be few established beliefs. Without direct evidence
of WOM effect, inferences have been made from experimental work
on the impact of positive and negative information. It is well
established that negative information usually has more impact on
judgment than positive information (Skowronski & Carlston,1989) but
this finding may not extend to the relative impact of PWOM and
NWOM on brand choice in familiar categories.
Although there is little evidence, it appears that marketers believe
that NWOM has more impact than PWOM. For example, Assael (2004)
states, “Negative word of mouth is more influential than positive word
of mouth” (though this claim may conflate relative incidence and
relative impact). Conventions in media publicity also support the idea
that negative information is more potent. According to the Kroloff
(1988) principle, negative copy is four times as persuasive as positive
copy.
When direct observation is not feasible, we have to ga ther
evidence on the relative impact of PWOM and NWOM using indirect
methods. One method is to measure Internet postings about brands
and th eir subseque nt sales performance (e .g., Godes & Mayzlin,
200 4). A problem with t his method is that there may be little
correspondence between the content of consumer-generated media
and face-to-face advice. One is not necessarily typical of the other,
and t he large amount of face-to-face advice is likely to be t he
dominant influence on consumption. Keller and Fay (2006) found
that 8% of advic e was Web mediated, 70% was face-to-fac e, and 19%
was by telephone. For th is reason, we did not specifically explore
the effect of Internet advice, though growth of Internet use is likely to
make this an increasingly important form of WOM. A second method
is to use laboratory experiments to investigate the response to
information on familiar brands. Other techniques that may be used
include role-play experiments a nd surveys. These methods also
present problems. Role-play may not typify naturally occurring
behavior, and the measures of PWOM and NWOM in surveys may be
subject to different degrees of bias that will disto rt the es timation of
their relative impact.
Since no single method can provide conclusive evidence, we adopt
a three-pronged approach designed to build a persuasive argument
about the relative impact of PWOM and NWOM. First, using both role-
play experiments and surveys, we find that PWOM usually has
somewhat more effect than NWOM. This finding is similar to
experimental evidence that positive and negative information have
much the same impact on attitudes when the brands are familiar
(Ahluwalia, Burnkrant, & Unnava, 2000; Ahluwalia, 2002). Second, we
describe how th e pre-WOM probability of purchase (hereafter
referred to as PPP) and other variables contribute to the impact of
WOM. We show that this evidence suggests that PWOM and NWOM
are closely similar behaviors, making it less likely that measures of the
two are subject to strongly differential bias. Third, we explain why
PWOM could have m ore effect than NWOM if the pre-WOM
probability of purchase (PPP) is less than 0.5, and we
find that this is
so.
The organization of the paper follows this three-pronged approach.
The empirical work is preceded by reviews of relevant research and
followed by a discussion of findings.
2. Previous research
2.1. Rarity and room for change
Fiske (1980) observed that negative information is usually rarer
than positive information and argued that this made negative
information more useful (or diagnostic) than positive information
because the latter could often be presumed. For example, evidence
that a brand is unreliable is more useful than evidence that the brand
is reliable because reliability may be assumed as the default condition
for modern products. Under these circumstances, we would expect
negative information to have more effect on judgment. Studies have
supported this “negativity effect” (e.g., Anderson, 1965; Chevalier &
Mayzlin 2003; Fiske, 1980; Mizerski, 1982; Mittal, Ross, & Baldasare,
1998).
Fiske's explanation may be expressed in terms of the gap between
the position implied by the message and the position held by the
receiver. Information that restates what the receiver believes may
increase certainty but is unlikely to change other aspects of a
receiver's judgment. In contrast, information that differs from the
receiver's position may change their judgment. In most circumstances,
the greater amount of positive information on everyday matters
ensures that the position of most receivers is positive so there will be
more impact from negative informat ion when it is received.
Exceptionally, when the receiver's expectation is negative and the
information received is positive, there could be a “positivity effect”.
Fiske's work was extended in the accessibility–diagnosticity (A–D)
theory of judgmental response (Feldman & Lynch, 1988; Lynch,
Marmorstein, & Weigold, 1988). These researchers also suggest that
negative information is more useful, or diagnostic, by virtue of its
rarity. According to the A–D theory, people use diagnostic information
in preference to more accessible information when both are available,
so that negative information should normally have dominance.
There are other explanations for the larger impact of negative
information. One of these is that the rarity of negative information
makes it surprising and, thus, draws more attention (Berlyne, 1954). A
second is the effect of attribution (Laczniak, DeCarlo, & Ramaswami,
2001; Mizerski, 1982). For example, a positive Web comment may be
discounted because of suspicion that it is “arranged”. Research on the
negativity effect is reviewed in detail by Skowronski and Carlston
(1989).
Psychological studies typically measure the way attitude is
changed by advice. In research on the purchase of brands, it is more
relevant to measure the change in the probability of purchase brought
about by advice. Accordingly, we measure the impact of WOM on
choice as a shift in the stated probability of purchase, from PPP to post-
WOM. If the PPP is below 0.5, there is more room for change in
response to PWOM than in response to NWOM. For example, if the PPP
is 0.4, PWOM can have a maximum effect of 0.6 (up to unity), whereas
NWOM has a maximum effect of 0.4 (down to zero).
2.2. Contrary responses to advice
In the preceding section, we assumed that PWOM makes a receiver
more positive and that NWOM makes a receiver more negative about
the object of advice. However, Laczniak et al. (2001) found that people
sometimes reacted against advice and became even more committed to
a brand that was subject to negative comment. This contrary response to
advice has also been observed by Wilson and Peterson (1989) and by
Fitzsimons and Lehmann (2004). Such contrary responses may be
explained by reactance theory (Brehm, 1966
). In Brehm's account,
reactance is a state of arousal that motivates the maintenance of self-
determination when it is under threat. Reactance can even occur when
people are directed to do things that they want to do.
Reactance effects can be strong in experimental work. In Fitzsimons
and Lehmann's study, the recommendation of an unattractive option
216 R. East et al. / Intern. J. of Research in Marketing 25 (2008) 215–224
shifted preference toward the attractive alternative choice while
recommendation of an attractive option reduced the preference for
this option. Thus, a bizarre feature of Fitzsimons and Lehmann's study
was that an attractive option was chosen more often when it was the
subject of negative advice than when it was the subject of positive
advice, making the advice counterproductive.
Contrary responses may be increased by the experimental design
used to investigate them. Advice in experiments is unsolicited and this
could arouse caution among participants. In addition, when the effects
of advice are measured immediately after receiving it, there is no time
for any reactance effects to fade. In retrospective surveys, measures
are taken some time after the claimed social exchanges have occurred.
Thus, survey evidence will show the eventual effect of the advice
rather than the immediate effect. This is more appropriate because
consumers do not usually purchase immediately after receiving
advice.
There are two other processes that could produce contrary
responses. One occurs when the receiver of advice disagrees with
the values of the advisor and exp ects to dislike what he/she
recommends and like what he/she dismisses. A second response
relates to “damning with faint praise”. If a receiver expects a strong
recommendation from a communicator but receives only a lukewarm
recommendation, the object's attraction may be diminished although
the advice is still positive. We are concerned about contrary responses
because, if these are common, it is difficult to interpret findings on the
impact of PWOM and NWOM.
2.3. Experimental studies on real brands: commitment and familiarity
Ahluwalia et al. (2000) propose that prior commitment to a brand
may prevent consumers from fully accepting useful negative informa-
tion about that brand. These researchers describe commitment as a
form of attitude strength and measure it with three items that relate
to brand retention. To measure the response to information, Ahluwalia
et al. use support arguments, counter-arguments, attitude, and
perceived diagnosticity (the subjective judgment of diagnosticity).
The first experiment conducted by Ahluwalia et al. showed that
high-commitment participants treated written negative information
about Nike (in the form of newspaper articles) as less diagnostic than
low-commitment participants. In addition, they gave this information
less weight and mobilized a substantial number of counter-arguments
compared with low-commitment participants. These findings were
confirmed in a second study that used a less popular brand of athletic
shoe.
In further work, Ahluwalia (2002) compared responses to written
positive and negative information on a brand when participants were
familiar or unfamiliar with the brand. When the brand was unfamiliar,
the negative information elicited more supporting arguments and was
perceived to have more diagnosticity and weight. When the brand was
familiar, there were no significant differences in the impact of positive
and negative i nformation. Thus, Ahluwalia argues that brand
familiarity attenuates perception of the greater diagnostic value of
negative information and suggests that, under these circumstances,
positive information may be perceived to be more diagnostic than
negative information when, objectively, this is not so. This work
follows earlier work by Wilson and Peterson (1989) and Sundaram
and Webster (1999), which showed that the impact of advice was
greatly reduced when the object of the advice was familiar.
If these experimental findings apply to WOM, we would expect
PWOM and NWOM to have similar impacts when the categories are
familiar. However, differences between the laboratory and the natural
setting may weaken inferences from one to the other and Ahluwalia
and her colleagues are careful not to claim that their findings can be
generalized to WOM.
In social psychology, the generalization of experimental findings to
natural settings has been a matter of concern for many years. Early
work in this field was done by Campbell (1957) and, more recently, by
Shadish, Cook, and Campbell (2002). The experimental work reviewed
above differs from our approach in several ways, particularly:
•
The short interval between exposure to the stimulus and measure-
ment of the response gives no time for the impact of information to
fade or develop. In surveys, there is usually a substantial interval
between the occurrence of WOM and the measurement of any effect.
• Ahluwalia and her colleagues use cognitive and attitudinal measures
to assess impact. We use a measure of purchase likelihood that, if
accurate, relates to sales gained or lost
3
. Attitude measures could
show an increase even when a person was fully committed to
repurchasing a brand and purchase likelihood was unchanged.
2.4. What causes WOM?
We discuss the antecedents of WOM because these cast light on
the nature of PWOM and NWOM. Our argument is that PWOM and
NWOM are similar behaviors, except for their opposed effects on
brand purchase. Researchers have claimed that PWOM is based on
satisfaction and NWOM on dissatisfaction (e.g., Goldenberg, Libai,
Moldovan, & Muller, 2007; Richins, 1983), which may provide a basis
for differential effects. However, Mangold, Miller, and Brockway
(1999), found that the satisfaction or dissatisfaction of the commu-
nicator and receiver are the catalysts of WOM in only 12% of cases.
Furthermore, PWOM and NWOM had the same triggers, which
occurred at similar frequencies. This indicates that the two forms of
WOM are very similar in origin. In Mangold et al.'s work, most PWOM
and NWOM arose as a response to the perceived need of another or
occurred as part of a conversation.
Turning to impact, if we can show that the impacts of PWOM and
NWOM rest on the same factors to the same degree, we will support
our argument that PWOM and NWOM are similar in nature and,
therefore, that biases in their measurement are likely to be similar.
2.5. Factors that may be associated with impact
Several factors may be associated with the impact of WOM.
1. Room for change. As we have noted, room for change in purchase
probability (in the direction indicated by the WOM) is limited by
the PPP, which could favor the impact of either PWOM or NWOM,
depending on the mean value of the PPP.
2. The strength of expression of WOM. It seems likely that the strength
of expression of WOM directly affects impact. However, sometimes
people react against advice, and this could produce an inverse or
more neutral relationship between strength of expression and
impact.
3. The closeness of the communicator to the receiver. Does advice from a
strong tie (a close friend or relative) have more impact than advice
from a weak tie (acquaintances and distant relatives)? Granovetter
(1973) argued that weak ties have more impact on the transmission
of information through a network of social groups because weak ties
tend to be members of more groups and can receive information in
one group and pass it on in another. Brown and Reingen (1987)
studied “who told whom” with regard to the customers of piano
teachers. Their findings supported Granovetter's thesis (network
effect), but they also found that receivers thought that strong ties
3
Purchase likelihood has a floor and ceiling, which limit change. Floor and ceiling
effects are often discussed as measurement problems. This is correct if the concept
does not have limits and the measure does. In this case, the limits (0, 1) are appropriate
for both concept and measure.
21 7R. East et al. / Intern. J. of Research in Marketing 25 (2008) 215–224
had more immediate impact (local effect). In this research, we test
the local effect.
4. Whether WOM is solic ited or not. East, Hammond, Lomax, &
Robinson (2005) studied the relative impact of solicited and
unsolicited WOM. They found that the impact ratio of solicited to
unsolicited WOM was approximately 1.5 to 1 for both PWOM and
NWOM. Bansal and Voyer (2000) also claimed that solicited WOM
has more impact, but neither of these studies took the effect of
other variables into account. It is possible that other variables, such
as the closeness of the communicator or the room for change, are
related to whether or not the advice is solicited. This could occur if
people ask for advice more often from those they are close to or
when there is more scope for change. The separate effects of these
variables can be established by using multiple regression analysis.
5. Whether the WOM is about the receiver's main brand. WOM about
the main brand may have a different effect than WOM about other
brands when factors such as PPP are controlled. We would expect
PWOM to be more acceptable and NWOM to be less acceptable
when it is about the main brand.
6. How much WOM the respondent reported giving on the category that
was studied. Here, although we had no prior evidence, it seemed
useful to know how those who give WOM are affected by the
receipt of PWOM and NWOM.
7. Age, gender and category. We tested the contribution of these
variables.
We did not directly measure the receiver's need for advice, which
would probably make a very substantial contribution to impact; our
view here was that this relationship would be close to tautological and
not very informative. We did not ask the receiver to assess the
communicator's satisfaction or dissatisfaction, which requires infer-
ence about the communicator's state of mind. To some extent, the
communicator's feelings will affect the strength of expression of
WOM, which can be directly assessed by the respondent. In addition,
we did not measure the receiver's stated brand commitment but
inferred this from the PPP.
3. Research questions
In summary, the questions that we raise are:
RQ1. Which has the most impact on brand choice, PWOM or NWOM?
RQ2. How do the variables identified above affect the impact of
PWOM and NWOM?
RQ3. Do the same variables explain the impact of PWOM and NWOM
to the same degree?
RQ4. How do the mean impacts of PWOM and NWOM relate to the
PPP?
RQ5. Does brand commitment reduce the impact of PWOM and
NWOM?
4. Research
4.1. Preliminary role-play experiments
4.1.1. Methods
The preliminary role-play experiments were conducted from 2003
to 2005 using eight surveys, seven in the UK and one in France. Non-
users of the category were excluded. All surveys were one-wave. No
incentives were used. All but one of the surveys covered two
categories, and some categories were covered more than once. We
present each category in a survey as a separate study, giving a total of
15 studies. The methods of questionnaire distribution are shown in
Table 1 together with the sample sizes and response rates (column 1).
The main method of distribution was house-to-house delivery, with
return by pre-paid mail. In each survey, the investigator delivered to
several middle-income suburban districts. A letter requesting help
accompanied the questionnaire. In some cases, the investigator spoke
to the householder, which may have raised the response rate. There
were 1905 respondents in total. The questionnaires carried a range of
questions, bu t only two were relevan t to the role-play study.
Respondents were asked to state how they would respond if they
received symmetrically phrased positive and negative advice from a
friend. The response was registered on a 7-point scale, and the
questions and scales are illustrated in Appendix A.
4.1.2. Findings
Table 1 shows the findings from the 15 studies on the effect of
positive and negative advice. Mean responses for PWOM and NWOM
are shown in columns 3 and 4, and the data are ordered by column 3.
In answer to RQ1, we find that the impact of the positive and negative
advice is similar but that PWOM has slightly more effect, which, with
the large amount of data available, is significant (p b 0.001, one-tailed,
Wilcoxon exact test). To check the influence of response rate, we
correlated the response rates with the scores in columns 3 and 4.
These correlations were not significant (p = 0.81; p = 0.93).
Table 1
Respondent judgments of influence on brand preference
Category (method
a
, sample size, response rate) Number available for analysis Judged impact (1–7)
Positive Negative
1 234
Cell phone airtime (DP, 170, 48%) 81 2.38 2.69
ISP (DP, 170, 48%) 73 2.45 2.53
Cell phone airtime (DP, 302, 39%) 113 3.73 3.21
ISP (DP, 302, 39%) 93 3.74 3.26
School (DS, (865, 14%) 122 4.31 3.48
Grocery store (France) (DP, 300, 59%) 173 4.45 3.98
Fashion store (France) (DP, 300, 59%) 173 4.57 4.32
Educ. institution (DPB, 665, 34%) 64 4.64 4.44
Cell phone airtime (DPB, 665, 34%) 190 4.74 4.97
Cell phone airtime (DP, 43%) 165 4.88 4.84
Credit card (DP, 400, 43%) 140 4.90 5.10
Optician (DPB, 665, 34%) 150 5.07 5.35
Optician (DP +DI, 280, 63%) 87 5.07 5.13
Coffee house (DP, 400, 43%) 104 5.14 5.00
Restaurant (DP+ DI, 280, 63%) 177 5.25 3.84
Means (unweighted) 127 4.35 4.14
a
Methods of gathering data: DP is drop-off with free post back; DS is via schools; DPB is distribution by paperboys; DI is face-to-face distribution and collection by intercept.
218 R. East et al. / Intern. J. of Research in Marketing 25 (2008) 215–224
4.2. Retrospective surveys
4.2.1. Methods
We surveyed the recalled impact of PWOM and NWOM on brand
purchase using 11 new surveys conducted from 2005 to 2007. Five of
the surveys covered one category, and seven covered two categories.
We report this research as 19 separate studies. The methods of
questionnaire distribution are shown in Table 2, together with the
sample sizes and response rates (column 1). All studies were one-
wave, and no incentives were used. In 16 studies, questionnaires were
delivered by hand to middle-income homes in urban areas near to
London in the UK. The investigator collected the completed ques-
tionnaire by arrangement (usually two days later), or pre-paid mail
was used for return. In three of these studies, supplementary data
collection was made via intercept or friends. In one study, members of
the public were approached in a coffee shop (luxury brands). The
study of luxury leather goods was conducted by distributing
questionnaires to customers of two stores in Lebanon. The study of
Iranian restaurants was restricted to Iranians living in the London
area, and distribution was via friends. The Lebanese study on luxury
leather goods and the hair colorant study (conducted in Japan) were
restricted to women. In total, we gathered data from 2544 respon-
dents, but, since only a portion of these had received PWOM or NWOM
on the focal category, analyses were conducted on smaller numbers.
The questionnaires covered a range of issues, and the relevant
questions are shown for one category in Appendix B. In all cases,
respondents were asked if they had received positive and negative
advice in the last six months on any brand in a specified category. If
advice had been received, respondents were asked to state whether
the last instance of PWOM/NWOM had affected their brand choice (or
had affected their prospective brand choice when delayed purchase
was likely). This measure allowed us to compare the proportion of
respondents who claimed to have been affected by PWOM with the
proportion of respondents who claimed to have been affected by
NWOM.
In each of these studies, we used the Juster scale (shown in
Appendix B) to measure the probability of purchase before and after
receiving WOM (Juster, 1966). The Juster scale measures probability in
10% intervals, and Wright and MacRae (2007) have shown that it
tracks objective measures quite closely.
Other questions allowed us to e stabli sh: (1) h ow strongly
expressed the advice was, (2) whether the communicator was close
to or distant from the receiver, (3) whether the advice was about the
main brand, (4) whether the advice was sought or not, and (5) how
much advice on the category was given by the respondent. We also
noted age and gender.
4.2.2. Findings
Table 2 shows the results. Column 1 shows the category, method,
sample size, and response rate. Columns 2 and 3 show the numbers of
respondents who report having received PWOM and NWOM. Columns 4
and 5 show the percentages of these respondents claiming that the last
instances of PWOM and NWOM had affected their decision (the studies
are ordered by column 4). Columns 6 and 7 are the PPPs for PWOM and
NWOM, and columns 8 and 9 are the mean shifts in the probability of
purchase produced by the last instances of PWOM and NWOM.
The data at the base of columns 4 and 5 show the mean impacts of
PWOM and NWOM. Overall, 64% claimed that PWOM, and 48%
claimed that NWOM, affected their decisions. A Wilcoxon test on the
individual data forming these percentages is significant (p b 0.001,
one-tailed exact test), and the difference between the study means
(columns 4 and 5 of Table 2) also reaches significance (p = 0.014, one-
tailed exact test). Based on these results, PWOM is more influential
than NWOM. When impact is measured as the shift in purchase
probability, we see that PWOM produces a mean shift of 0.20 and that
NWOM produces a shift of − 0.11, making PWOM 76% more influential
than NWOM. When absolute numbers are te sted, PWOM has
significantly more impact than NWOM in the pooled data (p b 0.001,
one-tailed exact test) as well as across studies when columns 8 and 9
are compared (p = 0.028, one-tailed exact test). Thus, using both
Table 2
The impact of PWOM and NWOM on brand choice probability
Category (method, sample size, usable
response rate)
Number in sample
receiving
Percent claiming
effect on decision of
Probability of purchase % Shift in probability of
purchase
PWOM NWOM PWOM NWOM Prior to PWOM Prior to NWOM PWOM NWOM
1 23456 7 89
Supermarket (DP +DF, 300, 31%) 42 35 33 54 0.43 0.39 0.16 − 0.16
Cell phone airtime (2007) (DC, 300, 64%) 55 50 42 40 0.40 0.41 0.16 − 0.09
Cell phone handset (2007) (DC, 300, 64%) 71 64 45 39 0.50 0.42 0.08 − 0.19
Current bank account (DC, 250, 65%) 113 89 56 45 0.40 0.47 0.28 − 0.11
Camera (DP, 300, 34%) 71 52 59 48 0.45 0.38 0.01 − 0.17
Computer (DC, 220, 80%) 106 71 60 68 0.53 0.49 0.20 − 0.20
Cell phone airtime (2005) (DC, 250, 86%) 149 152 61 53 0.32 0.41 0.19 − 0.10
Main credit card (DC, 250, 65%) 83 70 63 50 0.37 0.48 0.28 − 0.17
Luxury brands (CS, 115, 87%) 72 36 64 44 0.38 0.20 0.12 − 0.06
Leather goods, Lebanon (DS, 235, 74%) 166 159 65 34 0.48 0.46 0.23 − 0.14
Camera (DP, 300, 27%) 43 18 65 44 0.53 0.34 0.17 − 0.12
Holiday destination (2006) (DP, 300, 27%) 56 34 66 62 0.48 0.42 0.18 − 0.19
Coffee shop (DC, 220, 80%) 92 68 67 43 0.54 0.42 0.19 − 0.11
Holiday destination (2007) (DP, 300, 34%) 88 54 67 69 0.41 0.38 0.06 − 0.06
Cell phone handset (2005) (DC, 250, 86%) 157 155 70 35 0.39 0.36 0.20 − 0.07
Restaurant, favorite (DP+ DF, 300, 31%) 67 37 72 86 0.35 0.59 0.39 − 0.47
Restaurant, ethnic (DP +DI, 300, 30%) 75 43 73 86 0.36 0.41 0.34 − 0.23
Hair colorant (DC, 222, 77%) 45 18 78 39 0.51 0.28 0.19 − 0.08
Restaurant, Iranian (DF, 20 0, 45%) 79 58 86 43 0.44 0.22 0.31 − 0.03
Totals 1630 1263
Means (weighted) 64 48 0.43 0.40 0.20 − 0.11
Methods of gathering data: DP is drop-off with free post back; DF is distribution via friends; DC is drop drop-off and collect; CS is distribution and collection in coffee shop; DS is
distribution in stores; DI is face-to-face distribution and collection by intercept.
219R. East et al. / Intern. J. of Research in Marketing 25 (2008) 215–224
measures of impact, we get the same answer to RQ1: overall, PWOM
has more impact than NWOM.
We also conducted a check on the level of contrary responses
(where PWOM produces a negative shift or NWOM produces a
positive shift). Four percent of the responses to PWOM were negative,
and 7% of the responses to NWOM were positive. We did not find that
the omission or reversal of contrary responses increased the R
2
in later
regression analyses, and we used all responses as given.
RQ2 concerned the contribution of different variables to impact.
Step 1 was to use all the variables in an ordinal regression analysis to
predict the impact of WOM, m easured as a shift in p urchase
probability
4
. This produced a Cox and Snell R
2
of 0.30 for PWOM
and 0.28 for NWOM. Most of the category dummies made a significant
contribution, and the categories that related to the impact of PWOM
also related to the impact of NWOM (the correlation between
coefficients for the categories was 0.63, p = 0.005). Age and gender
were not significant in the regression analysis, and we do not further
consider these factors further.
Step 2 was to assess the effect on impact of the six variables
remaining after excluding age, gender, and categories. For this, we
used OLS regression. Initially, we included response rate in the
analyses to check its effect but it was found to be insignificant and was
removed. Table 3 is based on only 14 of the 19 studies because data
were missing on whether the advice was about the main brand in five
of the studies. Comparisons between the 19 and 14 studies, omitting
the advice about the main brand, indicated that the 14 studies were
typical of the 19. The variables in Table 3 are arranged in order of the
PWOM betas.
From Table 3, we see that the PPP, indicating room for change, has
the greatest beta weight for both PWOM and NWOM, followed by
strength of expression and whether the WOM was about the main
brand. These are the main contributors to impact among the variables
measured.
The closeness of the communicator does not reach significance for
NWOM in the regression analysis but a simple cross-tabulation
showed that shifts in the probability of purchase for PWOM and
NWOM are 32 and 51% greater for close ties compared to distant ties
(both significant, p b 0.001). Thus, it appears that the stronger impact
of close ties in a cross-tabulation depends partly on interactions with
other variables, the effect of which is removed in a multiple regression
analysis. In a similar manner, we tested the simple association
between impact and whether the advice was sought/uns ought.
Impacts for PWOM and NWOM were 24 and 17% greater when advice
was sought. The first was significant (p b .001), but the second did not
reach significance (p = 0.067). Here, again, it appears that the simple
association depends in part on other variables since sought advice did
not have a significantly greater effect than unsought advice in the
regression analysis. The last variable was the amount of WOM given.
This is significant in the case of NWOM, which means that those who
give more NWOM are more responsive to NWOM received
5
.
We turn now to whether PWOM and NWOM are determined by
the same variables to the same degree (RQ3). Table 3 shows that the
beta coefficients are similar except for whether the WOM was about
the main brand, where a reverse in sign is seen. This was expected
since it relates to the direction of effect of PWOM and NWOM. PWOM
goes with brand commitment but NWOM has an uphill task
dissuading a committed respondent. The correlation between the
two columns of beta weights in Table 3 (using absolute numbers) is
0.96 (p = 0.003), showing a high degree of symmetry.
RQ4 was about the relationship of the PPP to the mean impact of
PWOM and NWOM. Table 4 shows the relevant data and Fig. 1
presents these data in a more accessible form. For both PWOM and
NWOM, we see a relatively straight section on each plot that then
deflects toward the x-axis. These deflections can be attributed to the
effect of brand commitment and show how this factor constrains
impact.
We can compare the overall impacts under these conditions by
summing the scores on the y-axis for each point on the x-axis. This
gives PWOM a score of 1.94 compared with 1.56 for NWOM. So, PWOM
remains 24% more influential when the effect of the distribution of the
PPP is removed. Without removing the effect of the PPP distribution,
PWOM is 76% more influential than NWOM, indicating that about
two-thirds of the greater impact of PWOM can be related to the
greater room for change created by the distribution of the PPPs.
The plots in Fig. 1 provide a partial answer to RQ5 about the way
brand commitment reduces the impact of PWOM and NWOM. We can
estimate this effect numerically by imposing regression lines on the
last seven points of the PWOM plot and the first seven points of the
NWOM plot, and extending these over the areas of deflection. Without
the deflection, the score for PWOM would be 2.41 (24% more) and the
score for NWOM would be 1.97 (26% more). However, confining the
commitment effect to the relevant four points gives 44 and 46% more,
respectively.
It is possible that the deflections are related to other factors than
brand commitment. To test this, we compared the plots for high-
commitment categories against those for low-commitment cate-
gories. We reasoned that repertoire categories such as restaurants are
low-commitment because a consumer can easily include a new
restaurant in his/her repertoire. Also, dropping a resta urant in
response to NWOM is easier when there are several alternatives.
Based on these assumptions, we grouped restaurants (three studies),
leather goods, luxury brands, holiday destinations (two studies),
coffee shops, hair colorants and supermarkets as low commitment.
Fig. 2 shows the plots when the categories are divided in this manner.
While near-linear parts of the distribution are similar for the plot
4
Although the output measure is close to ratio-scale and OLS regression is probably
more appropriate, ordinal regression (in SPSS) has a convenient facility for creating
dummy variables (for the categories).
Table 3
Variables related to impact (NWOM impact treated as positive)
Variable PWOM (N =1108) NWOM (N = 903)
Beta S.E. t Sig. Beta S.E. t Sig
PPP 0.43 .024 15.7 b .001 0.37 .022 12.0 b .001
Strength of expression of WOM 0.22 .070 8.3 b .001 0.22 .06 5 7.3 b .001
WOM about main brand 0.16 .130 5.8 b .001 − 0.21 .164 − 6.6 b .001
Closeness of communicator 0.10 .120 3.8 b .001 0.06 .121 1.9 0.058
Whether advice was sought 0.06 .127 2.2 0.028 0.04 .140 1.4 0.167
Amount of PWOM/NWOM
given
0.04 .025 1.5 0.130 0.08 .022 2.8 0.005
Adjusted R
2
0.23 0.21
5
Binary interaction terms combining the strength of expression, closeness of the
communicator, whether the advice was about the current brand and whether the
advice was sought were added to the analyses and did not show significant beta
weights in the prediction of the impact of either PWOM or NWOM. Interaction terms
combining other variables with PPP were tested. In the prediction of PWOM, strength
of expression and current brand interactions with PPP were significant and raised the
R
2
from 0.23 to 0.26. In the prediction of NWOM, strength of expression, current brand,
closeness of the communicator and whether advice was sought were significant and
raised the R
2
from 0.21 to 0.26. This evidence shows that the impact of other variables
is moderated by room for change, as would be expected.
220 R. East et al. / Intern. J. of Research in Marketing 25 (2008) 215–224
pairs, the rest of the plot shows more deflection for the high-
commitment categories.
5. Discussion
5.1. Main findings
We examined the relative impact of PWOM and NWOM on
reported brand purchase probability. We used two methods of
investigation, employed three measures of impact in total, and
gathered data across a range of categories. Overall, PWOM had more
impact on brand purchase probability than NWOM. This finding is
similar to the findings from experiments. We have also shown that the
same determinants govern the impact of PWOM and NWOM, with
closely similar weights. This finding suggests that these two forms of
WOM are similar behaviors and, thus, that they are likely to have
similar measurement biases.
We provided an explanation for the greater effect of PWOM. We
found that the prior probability of purchase (PPP) tends to be below
0.5, leaving more room for change in response to PWOM than NWOM.
We also found that room for change is related to impact. We can thus
employ Fiske's gap explanation to explain why, in this case, there is a
“positivity effect”, with PWOM having more impact than NWOM. This
account differs from that of Ahluwalia (2002), who argued that
negative information was more diagnostic, but that brand familiarity
attenuated the perception of this diagnostic value. By interpreting
diagnosticity as the gap between receiver position and message, we
found that NWOM is less diagnostic than PWOM with regard to brand
choice.
5.2. Other findings
Our work also shows the effect of other factors on the impact of
PWOM and NWOM. The strength of expression of WOM has a strong
effect on both PWOM and NWOM. If the WOM is about the receiver's
main brand, it has a positive effect when it is PWOM and a negative
effect when it is NWOM.
We found that the impact of PWOM and NWOM from close ties
was more significant in a cross-tabulation than in the multiple
regression analysis. This suggests that part of the effect of close ties
(e.g., as observed by Brown & Reingen, 1987) comes from associated
variables. Similarly, we found that advice that was sought tended to
have more effect than unsought advice in a cross-tabulation. This
finding was significant for PWOM but not for NWOM, but the effect
was significant for neither PWOM nor NWOM in the regression
analyses. This suggests that earlier work by East et al. (2005) and
Bansal and Voyer (2000) should be treated cautiously since this work
excluded the effect of other variables. However, the simple association
may be more relevant in practical application if associated variables
are likely to be invoked by the application.
5.3. Applications
5.3.1. Setting the record straight
It is our understanding that both academic and practitioner
marketers believe that NWOM has more impact on brand purchase
than PWOM. Our evidence indicates that this belief is mistaken.
Marketers need to purge their discipline of beliefs that are little more
than hearsay, particularly when they apply to one of the most
powerful influences on consumption. We need to have an under-
standing of how PWOM and NWOM exert impact on consumer
decision making if we are to conduct more focused research in this
field.
5.3.2. Predictions based on the PPP
Fig. 1 indicates how the effect of WOM is conditioned by the PPP
and commitment. Other forms of communication such as advertising
and direct marketing could be similarly conditioned. From the plots
we see that positive messages have more impact when the PPP is 0 to
0.5 and that negative messages have more influence in the range 0.5 to
0.9. Thus, the potential impact of WOM, and possibly other commu-
nication types, can be estimated from Fig. 1 if the PPP of a segment can
be assessed using purchase records or management judgment, for
example.
Fig. 2. Shift in probability of purchase (impact) as a function of PPP. Categories separated
into high high-commitment and low low-commitment groups.
Fig. 1. Shift in probability of purchase (impact) as a function of PPP.
Table 4
Mean shift in purchase probability as a function of PPP
PPP PWOM NWOM
N Mean shift N Mean shift
0.0 137 0.26 185 0.02
0.1 132 0.24 128 − 0.02
0.2 203 0.29 182 − 0.01
0.3 241 0.28 141 − 0.09
0.4 167 0.25 120 − 0.13
0.5 258 0.19 179 − 0.16
0.6 159 0.18 105 − 0.25
0.7 124 0.12 92 − 0.26
0.8 116 0.06 80 − 0.28
0.9 57 0.02 50 − 0.22
1.0 64 − 0.06 61 − 0.13
Totals 1658 1.94 1323 − 1.56
221R. East et al. / Intern. J. of Research in Marketing 25 (2008) 215–224
5.3.3. A metric for WOM
Our work describes a method of measuring WOM impact using the
Juster scale, while a companion paper by East, Hammond and Wright
(2007) measures the relative incidence of PWOM and NWOM. By
combining the two methods, we obtain a metric for the net effect of
PWOM and NWOM and this may be compared with alternative
metrics. The main alternative is Reichheld's (2003) Net Promoter
Score (NPS). The NPS was designed to measure the number of people
who are likely to provide positive comments about the brand
(promoters) minus those likely to give negative comments (detrac-
tors). Many firms now use this measure to assess performance. While
Reichheld showed support for the NPS in a correlational study,
subsequent tests were less encouraging. Morgan and Rego (2006)
constructed a measure similar to the NPS and found it to be less
effective at predicting company revenue. Further work by Keining-
ham, Cooil, Andreasson and Aksoy (2007), using the industries studied
by Reichheld, found again that the NPS gave a poor prediction of
performance. In another study, Keiningham, Cooil, Aksoy, Andreassen
and Weiner (2007) showed that a multiple-item measure, rather than
the single-item NPS, gave a better prediction of retention and
recommendation.
This poor performance of the NPS may be because WOM is not
related to brand performance or because Reichheld's metric fails to
accurately measure the effect of WOM. We take the latter position. We
identify four potential weaknesses in the NPS in addition to any deficit
due to its single-item form. The first weakness is the use of self-
prediction (how likely is it that you would recommend … to a friend?)
since respondents cannot easily anticipate the circumstances that
would permit them to give advice about specific firms. The second
weakness is the ‘one size fits all’ feature of the NPS. The measure uses
an 11-point scale to measure the likelihood of recommendation,
assigning scores of 0 to 6 to detractors and scores of 9–10 to promoters
(the NPS is the difference in the percentages of respondents in these
segments). This measure does not allow for variation in the impact of
PWOM and NWOM across brands in a category. Third, NWOM is not
measured in the NPS but is inferred from low PWOM. However, those
who give little PWOM may not give NWOM. Indeed, East et al. (2007)
found that those who gave less PWOM also gave less NWOM and that
most of the NWOM given was on brands other than the focal brand.
East et al. (2007) found that the incidence of NWOM was more
variable in relation to market share than PWOM. This means that
specific brands may get more NWOM than PWOM even though there
is more PWOM than NWOM in the category. This makes it important
to measure NWOM as accurately as possible. The fourth weakness is
that the NPS measures the propensity to give advice but the advice
that people claim to have received is closer to impact on brand choice.
In our procedure, the incidence and impact of received PWOM and
NWOM on all brands in the category are measured. Using the product
of incidence and impact measures, we can assess the combined effect
of PWOM and NWOM on each brand in a category. We illustrate this in
Table 5 using our own data from 2005 on cell phones in the UK. Table 5
shows the brands, the number of instances of PWOM and NWOM
about each brand, the mean impact on brand purchase probability
that each instance of PWOM/NWOM produced, and the products of
number and impact. We combine the products to get the net effect of
PWOM and NWOM, which should relate to volume gain. To assess the
proportionate effect on market share, we divide the net effect by the
market share revealed by the respondents. Table 5 indicates that
Motorola and Samsung receive more support from WOM than other
brands but, with so few respondents, this is unlikely to be predictive.
This metric may perform better than the NPS and thus show that
WOM does predict brand performance.
6. Conclusion
We used role-play experiments and survey methods and found
that PWOM usually had more effect than NWOM. We explained why
this was found, adapting the explanation that is often cited in support
of the belief that NWOM has more impact than PWOM. We showed
that the impact o f both PWOM and NWOM had the same
determinants with closely similar beta weights, which suggests that
these two forms of WOM are similar behaviors. This makes it less
likely that our findings are distorted by differential recall bias. In this
way, we present a persuasive case that PWOM usually has more
impact than NWOM.
Acknowledgements
The contributions of the editor and reviewers greatly improved
this paper. Very useful advice was received from John Lynch. We
gratefully acknowledge those whose work was used in the preparation
of this paper: Chantal Adaimy, Pavadee (Sai) Chokesirikulchai, Jean-
Francois Damais, Steve Deschildres, Jo Eskell, Francesca Fanshawe,
Alman Gaba, Menekse Guven, Caroline Hancock, Monika Holbrack,
Onsitang Honda, Gelareh Hooshyar, Massa Iwata, Tehmina Jifri, Dilip
Joseph, Laprasada (Ja) Laksanasopin, Justin Sadaghiani, Siti Salwah
binte haji Saim, Kathryn Shirley, Omer Soomro, Lindsey Tregurtha,
Marike van Iersel, Iosif Vourvachis, and Dongjiao Xu.
Appendix A. Sample questions used in the preliminary studies
To what extent do you agree/disagree with the following? “Iam
looking for a new restaurant. A friend tells me that he/she has had a
negative experience with a restaurant. This would stop me from
going there”
Strongly disagree [1]
Disagree [2]
Slightly disagree [3]
Neither disagree nor agree [4]
Slightly agree [5]
Agree [6]
Strongly agree [7]
To what extent do you agree/disagree with the following? “Iam
looking for a new restaurant. A friend tells me that he/she has had a
positive experience with a restaurant. This would get me to go there”
Strongly disagree [1]
Disagree [2]
Slightly disagree [3]
Neither disagree nor agree [4]
Slightly agree [5]
Agree [6]
Strongly agree [7]
Table 5
Mobile phones: WOM effect: measures for 100 respondents
Brand N
P
Mean shift
P
N
P
×Mean shift
P
N
N
Mean shift
N
N
N
×Mean shift
N
Net effect Market share Proportionate effect
Nokia 147 0.17 25 99 − .08 − 8 17 40 0.43
Sony Ericsson 99 0.21 21 104 − .07 − 7 14 25 0.56
Motorola 94 0.19 18 55 − .06 − 3 15 14 1.07
Samsung 40 0.25 10 38 − .03 − 1 9 10 0.90
Siemens 11 0.22 2 38 − .05 − 2 0 4 0.00
Others 8 0.57 5 104 − .09 − 9 − 47 − .57
222 R. East et al. / Intern. J. of Research in Marketing 25 (2008) 215–224
Appendix B
In this questionnaire, we sometimes ask you to judge the likelihood of doing something from 0 to 10. Please rate your answers to these
questions according to the following scale:
10 Certain, practically certain (99 in 100)
9 Almost sure (9 in 10)
8 Very probable (8 in 10)
7 Probable (7 in 10)
6 Good possibility (6 in 10)
5 Fairly good possibility (5 in 10)
4 Fair possibility (4 in 10)
3 Some possibility (3 in 10)
2 Slight possibility (2 in 10)
1 Very slight possibility (1 in 10)
0 No chance, almost no chance (1 in 100)
1. Do you own a mobile phone? 11. In the last six months, how many times have you
No [1] received negative advice about any mobile phone
Yes [2] handset?
2. Which make of mobile phone do you have? Write in number (0, 1, 2 etc ……)
Have no mobile phone [1] If you answered 0, then please go to Q.19
Nokia [2] 12. The last time you received negative advice, did you
Sony Ericsson [3] ask for advice or was it just given?
Motorola [4] Just given [1]
Samsung [5] Asked for it [2]
Siemens [6] 13. What was your relationship to the person who last
Panasonic [7] gave negative advice?
NEC [8] Casual acquaintance [1]
Airtime supplier phone (O2, 3 etc) [9] More distant family, friend or colleague [2]
Other brand of mobile phone [10] Close family, close friend or colleague [3]
3. In the last six months, how many times have you 14. About which brand was the last negative advice
received positive advice about any mobile phone received? Please write in the make of mobile phone
handset? (Nokia, Sony Ericsson etc) ……………………………
Write in number (0, 1, 2 etc ……) 15. Did the last negative advice received affect your
If you answered 0, then please go to Q.11 handset choice or intended handset choice?
4. The last time you received positive advice, did you No [1]
ask for the advice or was it just given? Yes [2]
Just given [1] 16. From 0 to 10, how likely were you to choose the
Asked for it [2] handset before you received the last negative
5. What was your relationship to the person who last advice? (Please see the scale above)
gave you positive advice? Write in number (0 to 10)……
Casual acquaintance [1] 17. From 0 to 10, how likely were you to choose the
More distant family, friend or colleague [2] handset after you received the last negative advice?
Close family, close friend or colleague [3] (Please see the scale above)
6. About which brand was the last positive advice Write in number (0 to 10)……
received? Please write in the make of mobile phone 18. How strongly expressed was the last negative
(Nokia, Sony Ericsson etc) …………………………… advice?
7. Did the last positive advice that you received affect Hardly at all strongly [1]
your handset choice or intended handset choice? Moderately strongly [2]
No [1] Fairly strongly [3]
Yes [2] Very strongly [4]
8. From 0 to 10, how likely were you to choose the 19. In the last six months, how many times have you
handset before you received the last positive given negative advice about any mobile phone handset?
advice? (Please see the scale above) Write in number (0, 1, 2 etc ……)
Write in number (0 to 10) …… If you answered 0, then please go to Q.21
9. From 0 to 10, how likely were you to choose the 20. About which brand did you last give negative
handset after you received the last possible advice? advice? Please write in the make of mobile phone
(Please see the scale above) (Nokia, Sony Ericsson etc) ……………………
Write in number (0 to 10)…… 21. In the last six months, how many times have you
10. How strongly expressed was the last positive given positive advice about any mobile phone handset?
advice? Write in number (0, 1, 2 etc ……)
Hardly at all strongly [1] If you answered 0, then please go to Q.x
Moderately strongly [2] 22. About which brand did you last give positive
Fairly strongly [3] advice? Please write in the make of mobile phone
Very strongly [4] (Nokia, Sony Ericsson etc) ……………………
References
Ahluwalia, R. (2002, September). How prevalent is the negativity effect in consumer
environments? Journal of Consumer Research, 29,270− 279.
Ahluwalia, R., Burnkrant, R. E., & Unnava, H. R. (20 00, May). Consumer response to
negative publicity. Journal of Marketing Research, 37, 203−21 4.
Anderson, N. H. (1965, July). Averaging versus adding as a stimulus combination rule in
impression formation. Journal of Personality and Social Psychology, 2,1−9.
Arndt, J. (1967, August). The role of product-related conversations in the diffusion of a
new product. Journal of Marketing Research, 4,291−295.
Assael, H. (2004). Consumer behavior: A strategic approach. Boston: Houghton Mifflin
Company.
Bansal, H. S., & Voyer, P. A. (2000). Word-of-mouth processes within a services purchase
decision context. Journal of Service Research, 2(3), 166−177.
Berlyne, D. E. (1954). A theory of human curiosity. British Journal of Psychology, 45,
80−191.
Brehm, J. W. (1966). A theory of psychological reactance. New York: Academic Press.
In this questionnaire, we sometimes ask you to judge the likelihood of doing something from 0 to 10. Please rate your answers to these
questions according to the following scale:
223R. East et al. / Intern. J. of Research in Marketing 25 (2008) 215–224
Brown, J. J., & Reingen, P. H. (1987). Social ties and word-of-mouth referral behavior.
Journal of Consumer Research, 14(3), 350−362.
Campbell, D. T. (1957). Factors relevant to the validity of experiments in social settings.
Psychological Bulletin, 54,297−312.
Chevalier, J. A., & Mayzlin, D. (2003). The effect of word of mouth on sales: Online book
reviews. Journal of Marketing Research, 44 (3), 345−354.
East, R., Hammond, K., Lomax, W., & Robinson, H. (2005). What is the effect of a
recommendation? The Marketing Review, 5(2), 145−157.
East, R., Hammond, K. A., & Wright, M. (2007). The relative incidence of positive and
negative word of mouth: A multi-category study. International Journal of Research in
Marketing, 24(2), 175−184.
Feldman, J. M., & Lynch, J. G., Jr. (1988, August). Self-generated validity and other effects
of measurement on belief, attitude, intention, and behavior. Journal of Applied
Psychology, 73,421−435.
Fiske, S. T. (1980). Attention and weight in person perception: The impact of negative
and extreme behavior. Journal of Personality and Social Psychology, 38(6), 889−906.
Fitzsimons, G. J., & Lehmann, D. R. (2004). When unsolicited advice yields contrary
responses. Marketing Science, 23(1), 82−95.
Godes, D., & Mayzlin, D. (2004). Using online conversations to study word of mouth
communication. Marketing Science, 23(4), 545−560.
Goldenberg, J., Libai, B., Moldovan, S., & Muller, E. (2007). The NPV of bad news. Inter-
national Journal of Research in Marketing, 24(3), 186−200.
Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78(6),
1360−1380.
Juster, F. T. (1966, September). Consumer buying intentions and purchase probability:
An experiment in survey design. Journal of the American Statistical Association, 61 ,
658−696.
Keaveney, S. M. (1995). Customer switching behavior in service industries: An
exploratory study. Journal of Marketing,
59(2), 71−82.
Keiningham, T. L., Cooil, B., Andreasson, T. W., & Aksoy, L. (2007). A longitudinal
examination of ‘net promoter’ and firm revenue growth. Journal of Marketing, 71(3),
39−51.
Keiningham, T. L., Cooil, B., Aksoy, L., Andreassen, T. W., & Weiner, J. (2007). The value of
different customer satisfaction and loyalty metrics in predicting customer retention,
recommendation, and share-of-wallet. Managing Service Quality, 17(4), 361−384.
Keller, E., & Fay, B. (2006). Single-source WOM measurement: Bringing together senders
and receivers inputs and outputs. In W. J. Carl (Ed.), Measuring word of mouth, vol. 2
(pp. 31–41). Chicago: Word of Mouth Marketing Association.
Kroloff, G. (1988). At home and abroad: Weighing in, public relations journal. October, 8.
Laczniak, R. N., DeCarlo, T. E., & Ramaswami, S. N. (2001). Consumers' responses to
negative word-of-mouth communication: An attribution theory perspective. Jour-
nal of Consumer Psychology, 11(1), 57−74.
Lynch, J. G., Jr., Marmorstein, H., & Weigold, M. F. (1988, September). Choices from sets
including remembered brands: Use of rec alled attributes and prior overall
evaluations. Journal of Consumer Research, 15,169−184.
Mangold, W. G., Miller, F., & Brockway, G. R. (1999). Word-of-mouth communication in
the service marketplace. Journal of Services Marketing, 13(1), 73−89.
Mittal, V., Ross, W. T., & Baldasare, P. M. (1998, January). The asymmetric impact of
negative and positive attribute-level performance on overall satisfaction and
repurchase intentions. Journal of Marketing, 62,33−47.
Mizerski, R. W. (1982). An attributional explanation of the disproportionate influence of
unfavorable information. Journal of Consumer Research, 9(1), 301−31 0.
Morgan, N. A., & Rego, L. L. (2006). The value of different customer satisfaction and
loyalty metrics in predicting business performance. Marketing Science, 25(5),
426−439.
Reichheld, F. F. (2003). The one number you need to grow. Harvard Business Review, 81(12),
46−54.
Richins, M. (1983). Negative word of mouth by dissatisfied customers: A pilot study.
Journal of Marketing, 47(1), 68−78.
Shadish, W. R., Cook, T. T., & Campbell, D. T. (2002). Experimental and quasi-experimental
designs for generalized causal inference. Boston: Houghton-Mifflin.
Skowronski, J. J., Ca, & rlston, D. E. (1989). Negativity and extremity biases in impression
formation: A review of explanations. Psychological Bulletin, 105(1), 131−142.
Sundaram, D. S., & Webster, C. (1999). The role of brand familiarity on the impact of
word-of-mouth communication on brand evaluations. Advances in Consumer
Research, 26,664−670.
Wilson, W. R., & Peterson, R. A. (1989). Some limits on the potency of word-of-mouth
information. Advances in Consumer Research, 16,23−29.
Wright, M., & MacRae, M. (2007). Bias and variability in purchase intention scales.
Journal of the Academy of Marketing Science, 35(4), 617−624.
224 R. East et al. / Intern. J. of Research in Marketing 25 (2008) 215–224
